Peijun Liu1, Man Wang1, Yining Wang2, Min Yu3, Yun Wang1, Zhuoheng Liu3, Yumei Li1, Zhengyu Jin1. 1. Department of Radiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100730, China. 2. Department of Radiology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100730, China. Electronic address: wangyining@pumch.cn. 3. CT Business Unit, Neusoft Medical System Company, Shenyang, China.
Abstract
RATIONALE AND OBJECTIVES: To evaluate deep learning (DL)-based optimization algorithm for low-dose coronary CT angiography (CCTA) image noise reduction and image quality (IQ) improvement. MATERIALS AND METHODS: A postprocessing platform for the CCTA image was built using a DL-based algorithm. Seventy subjects referred for CCTA were randomly divided into two groups (study group A with 80 kVp and control group B with 100 kVp). Group C was obtained by DL optimization of group A. Subjective IQ was blindly graded by two experienced radiologists on a four-point scale (4-excellent,1-poor). The image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were calculated to evaluate IQ objectively. The difference between the time consumed of iterative reconstruction and DL algorithm was also recorded. RESULTS: The subjective IQ score of group C using the DL algorithm was significantly better than that of group A (p = 0.005). The noise of group C was significantly decreased, while SNR and CNR were significantly increased compared to group A (p < 0.001). The subjective IQ scores were lower in group A compared to group B (p = 0.037), whereas subjective IQ scores in group C were not significantly different (p = 0.874). For objective IQ, the noise of group A was significantly higher, while SNR and CNR were significantly lower than that of group B (p < 0.05). There was no significant difference in noise and SNR between group C and group B (p > 0.05), but CNR in group C was significantly higher than that in group B (p < 0.05). The DL algorithm processes the image twice as fast as the iterative reconstruction speed. CONCLUSION: The DL-based optimization algorithm could effectively improve the IQ of low-dose CCTA by noise reduction.
RCT Entities:
RATIONALE AND OBJECTIVES: To evaluate deep learning (DL)-based optimization algorithm for low-dose coronary CT angiography (CCTA) image noise reduction and image quality (IQ) improvement. MATERIALS AND METHODS: A postprocessing platform for the CCTA image was built using a DL-based algorithm. Seventy subjects referred for CCTA were randomly divided into two groups (study group A with 80 kVp and control group B with 100 kVp). Group C was obtained by DL optimization of group A. Subjective IQ was blindly graded by two experienced radiologists on a four-point scale (4-excellent,1-poor). The image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were calculated to evaluate IQ objectively. The difference between the time consumed of iterative reconstruction and DL algorithm was also recorded. RESULTS: The subjective IQ score of group C using the DL algorithm was significantly better than that of group A (p = 0.005). The noise of group C was significantly decreased, while SNR and CNR were significantly increased compared to group A (p < 0.001). The subjective IQ scores were lower in group A compared to group B (p = 0.037), whereas subjective IQ scores in group C were not significantly different (p = 0.874). For objective IQ, the noise of group A was significantly higher, while SNR and CNR were significantly lower than that of group B (p < 0.05). There was no significant difference in noise and SNR between group C and group B (p > 0.05), but CNR in group C was significantly higher than that in group B (p < 0.05). The DL algorithm processes the image twice as fast as the iterative reconstruction speed. CONCLUSION: The DL-based optimization algorithm could effectively improve the IQ of low-dose CCTA by noise reduction.